Highlights 2023
WP 23/02
In “Refining Public Policies with Machine Learning: The Case of Tax Auditing”, Luigi Guiso, together with Marco Battaglini, Eleonora Patacchini, Chiara Lacava and Douglas L. Miller, studies how ML techniques can be used to improve tax-auditing efficiency using administrative data without the need for randomized audits. It uses Italy's population data on sole proprietorship tax returns and audits and offers a new approach that addresses the challenge that predictions must be trained on human-selected data. The key finding is that there are substantial margins for raising revenue from audits by improving the selection of taxpayers to audit with Machine Learning. Replacing the 10% least promising audits with an equal number selected by the algorithm increases detected tax evasion by as much as 38%, and evasion that is actually paid back by 29%. Gains remain large even if the algorithm suggested audits are restricted to taxpayers in specific income groups or industrial sectors.
WP 23/01
In “How much and how fast do investors respond to equity premium changes? Evidence from wealth taxation”, Luigi Guiso, together with Andreas Fagereng and Marius Ring, investigates how individual investors respond to predicable changes in the equity premium. The paper exploits a wealth tax reform that first induced an equity premium and later reversed to track individual investor responses. Using administrative data on Norwegian investors’ portfolios, the paper documents strong but slow portfolio allocation responses to the equity premium shock. Tracking the portfolio allocations over time, it shows that the short-run responses are weak and resemble those found in a growing survey-based literature that uses variation in subjective beliefs of stock returns. However, the longer-run responses are much larger and can be rationalized by a contained coefficient of relative risk aversion between 2 and 3. The finding that response size increases significantly over time suggests that portfolio adjustment frictions are pervasive. The paper further finds that equity premium increases have large effects on stock market entry but that an equity premium decrease barely induces exits, which is consistent with the presence of both moderate entry costs, estimated at around $800, and considerably smaller per-period participation costs of around $90. The paper findings of slow portfolio responses to equity premium shocks provides supportive evidence for recent strands of literature that build on adjustment frictions to explain a wide range of asset pricing puzzles. They also have implications for optimal capital taxation when tax rates can differ across assets.
